- This change comes after long observations around how pipelines doesn't work well for copying large blocks of data mainly due to the segment allocation strategy. Today we allocate each segment > minimum segment size < max pool size. This works well if data is being quickly consumed because we can keep memory allocations to a minimum but doesn't work well when there's large chunks of data are being written. This change doubles the segment size based on the previous segment up to 1MB (arbitrary limit). This should make the default behavior work pretty much the same as today but performance of larger writes/reads should improve.